Printer Friendly

What you should know about web-inspection systems.

Today's quality-conscious film processors are using the camera's eye as a way to find the flaws that eat up profits

Film processors may need a certain kind of vision to thrive in today's age of heightened quality consciousness. But that vision increasingly may reside in a machine. Finding tiny flaws in web products has become too important a pursuit to leave to the vagaries of human inspection. For two decades, laser-based automatic web inspection has offered both consistency and on-line detection capabilities to the hunt for defects--both qualities lacking in manual inspection. But these relatively costly systems were the province of the "high-end" user whose line speeds and quality requirements needed the superior resolution of a laser sensor. More recently available machine-vision systems, by contrast, scan fast-moving webs with relatively inexpensive linear-array cameras. As film customers grow more exacting, camera-based inspection makes sense on a wider variety of webs than ever before.

Catching defects on-line means yield gains. Flow Vision says one of its customers, Exxon Corp., attributes to web inspection a 5-10% yield increase on an 80-in.-wide polyolefin line at the Lake Zurich, Ill., extrusion facility for specialty films. And some of Intec's customers have lines producing about $10,000 worth of product per hour, making real-time correction abilities extremely attractive, says Intec marketing v.p. Joseph Paumi.

All the systems discussed here position video cameras (charge-coupled devices or CCDs) above, or occasionally below, the web to collect light shining through or bouncing off the web. Computers and special software find defects by analyzing the transmitted or reflected light patterns from the web. Despite those similarities among machine-vision systems, picking the right one for your application is hardly cut-and-dried. Suppliers agree that the choice boils down to the type of defect to be found at a given line speed and product width. Those factors help determine considerations of cost and configuration. After all, someone looking for thumb-sized holes at 200 ft/min wouldn't have the same requirements--or incur the same costs--as a processor looking for subtle scratches at 1000 ft/min. In fact, suppliers universally prefer to test sample products in a laboratory setup before recommending a production system.

Of course, vision systems can't cure all quality ills. "CCD is a substitute for manual inspection but not sensitive enough to find the smallest defects at high line speeds," says Systronics applications consultant Brian Heil. In fact, all the suppliers acknowledge that laser-based systems still provide better resolution, enabling them to find smaller defects with more consistency at faster line speeds. Traditional laser systems also cost more--"multiple hundreds of thousands of dollars," Intec's Paumi says. At least two companies, however, have recently come out with laser systems priced competitively with camera systems.

Who can benefit from vision-based inspection systems and who should stick with lasers? CCD system suppliers say their customer base spans a wide range of applications. Film resin producers, for example, regularly check resin batches for propensity to cause gels. Film producers look for unsightly contaminants or gels. Extrusion coaters inspect both their own process and the film substrate they use. "Almost anyone whose customers are demanding better quality can benefit," says Flow Vision sales manager James Borges.

At the same time, CCD advocates do acknowledge some applications where line speeds and stringent defect requirements rule out a camera-based system. Systronics' Heil admits that "sometimes a camera just can't cut it." But he estimates that only about 10% of plastics applications present challenges that a camera can't meet more cost-effectively than a laser.


Even camera-based systems differ among themselves in terms of price and performance. But these differences have little to do with the front-end technology, which nearly everyone shares. These visual systems all detect web flaws with CCD cameras that project a linear array of light-sensitive pixels across the moving web. Sequentially reading information from the line of pixels, the cameras create a video signal. Defects alter the intensity of light shining through or bouncing off the moving web and thus turn up as signal discrepancies.

Some defects, like pinholes, would transmit more light, creating a jump in the signal relative to the product's "noise," or normal signal variations. On the other hand, a contaminant would block light, causing a temporary decrease in signal. Not everything is that simple. Sira project engineer Peter Campbell points out that many flaws affect light intensity in several ways at once. A hole, for instance, might scatter light around its edges but permit increased intensity through the middle.

Because flaws vary, so must lighting configurations. Lighting from behind the web may do the trick for common defects like holes and contaminants. Yet backlighting can't solve all detection problems. Heil points out that surface defects like streaks, voids, or dull spots show up better under a light reflected off the web rather than transmitted through it. And for finding gels in clear film, Systronics recommends "dark field" lighting. By aiming the linear array near but not directly at the light source, a signal would be produced only when a gel or other flaw scattered light into the camera's field of view. Intec's Paumi also notes that lighting configurations must accommodate some relatively subtle differences between flaws. Even the angle of reflected light can provide clues about a defect, often creating a need for "multiple viewing angles," he says.

Standard lighting on most systems comes from high-output fluorescent bulbs, though specialized light sources like fiber optics are available as needed. Flow Vision's Borges says that the "H-O" bulbs cost about $50 and last roughly 8-12 months.


The minimum flaw size a system can discover depends mostly on resolution. Yet in some measure, it also rests on individual beliefs about what constitutes a reliably detectable defect.

In the crossweb direction, the number of pixels projected onto the web determines the resolution. The most common cameras have 2048 pixels total, but suppliers offer arrays of 512, 1028 and 4096 as well. The resolution would equal the web width divided by the number of pixels a company considers usable. Flow Vision, for instance, employs 2000 pixels of a 2048-bit array, according to Borges. Sira feels only 1800 really count because some pixels overlap while others might be used for edge detection.

Anyone can bring more pixels to bear on a web by adding multiple cameras. Going to four CCD units--a common number for wider webs--would train about 8000 usable pixels on a web. But Intec's Paumi says that the camera-based systems tend to lose their pricing edge over lasers when resolution requirements call for eight or more cameras.

In the machine direction, the line speed and camera's scan rate determine how far the pixel projection stretches downweb. As line speed increases, so does "pixel stretch," decreasing the machine-direction resolution. For example, a static 60-in. film seen by 1800 usable pixels would have a downweb resolution of 0.033 in., calculates Sira's Campbell. But the resolution on that same web moving at 1000 ft/min would be 0.083 in.

Scanning more lines per second yields better downweb resolution. Common scan rates range between 2000 and 5000 lines/sec. As an option, some suppliers offer cameras with fewer pixels to scan as a way to attain even faster rates, but this increase must be balanced by the loss in crossweb resolution.

With resolution so easily changed by adding cameras or picking ones with a high scan rate, many systems can be configured to arrive at comparable degree of resolution. "No one CCD camera can do better than the rest," says Mayan Automation president Michael Braeuel. Still, minimum detectable defect size remains a matter of debate. Sira believes defect size should equal at least 100% of the projected pixel area to show up reliably. "If the defects are around 2-3 mils, there's no chance of seeing them in a projected pixel size 33 mils wide," Campbell says. And he notes that defects smaller than pixel area have another problem. They can straddle adjacent pixel projections without registering enough of a signal in either one to be counted consistently.

Alternatively, Flow Vision presents the problem in terms of signal-to-noise ratio. Defects should simply produce a signal two to three times greater than the background "noise" level to be considered consistently detectable, argues Borges.

So it all comes down to the expected size of the defects you want to detect. Heil explains that defects down to 1 mil have been caught by some Systronics multiple-camera systems. In actuality, though, the company's plastics customers more often want to find defects between 7 and 20 mils. Heil adds that this size can often be detected with CCD systems for half the price of a laser system. "Typically, we're looking for visual defects that would not compromise the product's performance," he explains.

CCD systems function satisfactorily at a variety of line speeds, but suppliers acknowledge limitations here, too. Heil cites around 600 ft/min as a typical speed for many customers' applications, but requirements vary widely. A recent Systronics installation looks for 0.1-in. defects at 2000 ft/min. "Laser systems just weren't practical for this larger defect," he says.


For some processors, merely finding defects doesn't do the trick. Instead, the demands of statistical process control (SPC) have engendered a need to classify flaws as well. "Characterization is the big push right now," says Paumi, who points out that each type of flaw may have its own distinct remedy.

To differing degrees, most web-inspection systems can identify classes of defect and provide frequency counts for each. They can also trigger an appropriate response like process changes, marking, or documentation.


Camera-based inspection systems may begin with shared technology, but some hardware differences still shine through. Fully installed turnkey installations command higher prices than do-it-yourself kits of components. Turnkey packages include lighting, camera lens, computers, installation and support. Their software offers defect mapping, analysis by each pixel's lanes on the web, trend graphs and historical reports. For those who simply want basic detection options, less expensive approaches rely on cameras with simpler controllers. So consider these sorts of features when looking at these newest offerings from these major suppliers:

* Ektron, a subsidiary of Eastman Kodak Co., recently began to market its Veredus Surface Flaw Analysis System for plastic film inspection. Until now it has been applied to Kodak paper and film as well as metal webs. Using up to eight 2048-bit cameras and scan rates up to 10,000 times/sec, the system works at line speeds up to 5000 ft/min. It can also provide two-sided inspection with cameras positioned above and below the web. Lighting comes from a proprietary linear source. Veredus employs a custom digital signal processor with a keyboard interface and gray-scale monitor. Systems include training and custom-designed modular installations. Depending on the options, a four-camera model would cost between $400,000 and $700,000, according to marketing manager James Meehan.

* Flow Vision offers the Advisor FDS Plus system for line speeds up to 1000 ft/min; it classifies up to four different types of defects at sizes down to roughly 2 mils. Scanning rates range between 2000 and 4000 times/sec. The system's industrial computer with color monitor displays defect mapping and trend analysis. According to Borges, a single-camera system complete with lighting and installation costs around $42,000. A top-of-the-line four-camera system would go for roughly $95,000.

* Futec International custom builds a proprietary line of CCD cameras with linear arrays of 256, 1024, 2592 and 5000 pixels. Futec reports scan rates of 12,000 times/sec for its 5000-pixel models. According to v.p. Yusoo Mii, the company's PKD system with two 5000-pixel cameras would cost about $75,000 and find 5-mil defects at 600 ft/min. A four-camera system would run about $100,000. Defect mapping can be displayed on a CRT terminal or personal computer.

* Harland Opposite's Auto-Site systems feature one to four cameras and scan rates of 10,000 times/sec. Defect maps appear on a color CRT. Prices start at $85,000 for a one-camera system.

* Intec's Series 9000 inspection system works with either CCD or laser technology. With CCD sensing, the system can use up to eight cameras at once. Scan rates run from 1200 to 9600 times/sec. Using logic-based definitions, the system can define up to 24 defect types for each stored product code. Defect maps of up to 200 lanes appear on the system's color monitor and optionally on a printed roll summary. Single-camera turnkey systems would start at $75,000 while eight-camera setups go for $250,000.

* Mayan Automation integrates a 2048-bit CCD array, lens and electronics in a single unit designed for easy mounting above a film line. Called the Fine-Line, this camera scans at 2000 times/sec. Its on-board electronics digitize and process scanned information immediately and directly signal any defects. Serial ports can also link the system to an optional PC. Menu-driven setup software is included. A FineLine alone costs $6000. While a complete system with light source, PC and training would go for roughly $15,000.

* Sira recently introduced a new automatic web-inspection system called the Series 10 that works with either laser or CCD sensors. Designed with user-friendliness in mind, it features Windows-based software and a touchscreen PC. The system can accommodate up to four cameras with scan rates of 2000-4000 times/sec. Its defect-mapping and trend-analysis capabilities can handle up to 15 defect classifications. As with all its systems, Sira offers remote support of the Series 10 through modem diagnostics. The Series 10 is priced around $80,000 for a one-camera turnkey system capable of finding 0.030-in. defects on a 60-in. web at 800 ft/min.

* Sunx Sensors sells the IX series of CCD cameras for a variety of tasks, including web inspection. Systems can include either 512-, 2048- or 4096-pixel cameras and a controller capable of storing up to 10 programs.

* Systronics offers systems using between one and eight 2048-pixel cameras with scan rates of up to 5000 times/sec. A turnkey four-camera system with six defect classes would cost $125,000. It includes a computer and touchscreen monitor, allowing operators to easily access defect maps and trend analysis. Service can be provided via modem. Also available from Systronics is a basic inspection system called The Eagle. For roughly $20,000, this self-installed system offers "basic defect detection" of gels and holes without the more sophisticated analytical software.

New Lower-Cost Laser Systems

For many inspection applications, camera systems are much less expensive than laser-based systems. They're also lighter and more easily installed. But inspection pros concede that CCD cameras just don't "see" as well as laser-based systems. Intec and Sira, which sell both lasers and CCD systems, have also come out with laser systems that may challenge some CCD-camera packages on one of their chief selling points--lower price.

Intec recently introduced a lower-cost laser system called the Series 3000. Joseph Paumi, marketing v.p., explains that it uses a solid-state infrared laser in place of the more common gas-laser approach. At roughly $100,000, a complete Series 3000 system costs well below some CCD systems. Sira, meanwhile, uses retro-reflective technology, whereby a single unit sends and collects the laser beam. It holds-down cost by eliminating the expense and need for a separate light-collecting device. At the same time, retro-reflective laser units provide separate channels for bright field detection of defects that intensify light and for dark field detection of defects that scatter light. A basic "low-cost" laser inspection system from Sira would also go for around $100,000.

When it comes to performance, laser sensing remains the touchstone for measuring an inspection system. Intec's Paumi says lasers commonly have crossweb resolutions down to 0.0016 in., while a camera system would normally bottom out around 0.002 in. Lasers may also be more consistent because they don't suffer from the partitioning of defects across pixel lines, which can create a misleading signal, says Paumi.

And Sira project engineer Robert Meszaros notes that the point light source from a laser travels the entire web, developing a signal from actual scanned product. Cameras, by contrast, derive their signals from a much larger area. Multiple cameras can increase resolution, but for the most rigorous resolution requirements, the balance shifts from a cost and performance perspective. Paumi says a single laser sensor is more cost-effective when the desired resolution would need eight cameras. Further shifting the balance, adds Meszaros, is the continuing trend to faster line speeds and tighter defect specifications.


Ektron, Rochester, N.Y. Flow Vision Inc., Wilmington, Mass. Futec International, North Brunswick, N.J. Harland Opposite Inc., New Haven, Conn. Intec Inc., Trumbull, Conn. Mayan Automation Inc., Lachine, Que. Sira Inc., Darien, Conn. Sunx Sensors, West Des Moines, Iowa Systronics Inc., Norcross, Ga.
COPYRIGHT 1992 Gardner Publications, Inc.
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1992, Gale Group. All rights reserved. Gale Group is a Thomson Corporation Company.

Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:includes related article; advantages offered by web-inspection systems over human manual inspection
Author:Ogando, Joseph
Publication:Plastics Technology
Date:Sep 1, 1992
Previous Article:What it takes to serve the automotive 'transnationals.' (tips on dealing with Japanese automotive transnationals)(includes related articles)
Next Article:Flame retardants update: phosphorus expands its niche.

Related Articles
ISO 9000 installation: keeping it in-house.
ISO 9002: ways to solve problem areas for foundries.
Audit and inspection tools for children's play areas.
Nondestructive testing: 5 ways to ensure defect-free deliveries.
Automatic inspection of surface imperfections and dimensions on rotational symmetric mass products.
Tools for Environmental Health.
Choosing an Automated Inspection Strategy -- AOI and AXI are viable and often complementary inspection technologies to use in PCB assembly.
Proper operation, maintenance, and servicing of aerobic wastewater treatment systems. (Guest Commentary).
AOI in a high-mix/low-volume environment: the cost savings when using AOI can pay for the system on one job under the right circumstances.
CDC and HUD to release an updated housing inspection manual--the Healthy Housing Reference Manual.

Terms of use | Copyright © 2018 Farlex, Inc. | Feedback | For webmasters